Re: [codec] I-D Action:draft-ietf-codec-opus-05.txt

[Koen Vos <[e-mail]>]

Hi Keith, 

Thanks for your report. 

The in-band FEC in OPUS Voice mode is a bit different from SILK's in that SILK keeps a separate full state of the noise shaping quantizer for the in-band FEC coding, whereas OPUS copies the state of the regular noise shaping quantizer before coding the FEC data. This reduces the OPUS state size by ~4 kB. There are other examples of where OPUS quality was slightly reduced in favor of complexity or memory. In my experience, packet loss above about 10% is typically of rather bursty nature, meaning that FEC won't be very helpful anyway. 

The good news is that the in-band FEC is still very effective at dealing with (random) packet loss, as suggested by the plot below (again WB speech coded at 40 kbps). 

I'll continue to look for ways to improve quality. 

best, 
koen. 



----- Original Message -----
From: "Keith Yan" <[e-mail]> 
To: "Keith Yan" <[e-mail]>, "Jean-Marc Valin" <[e-mail]> 
Cc: [e-mail] 
Sent: Tuesday, March 22, 2011 7:03:02 AM 
Subject: [codec] I-D Action:draft-ietf-codec-opus-05.txt 




Dear Jean-Marc, 

I have done some FEC related testing based on V0.9.3, from the test result, Opus can congest lost rate <10% well, but to high lost rate, it performances not as well as SILK. 

I do a competition with SILK V1.0.7. Following is my test condition: 

a. Set encode bitrates to 40kbps 

b. Enable inband FEC and set max lost rate is 20% 

c. Set OPUS_MODE_VOICE for Opus V0.9.3, MODE_SILK_ONLY for Opus V0.9.1, and set BANDWIDTH_WIDEBAND for V0.9.3 and V0.9.1. 



Test it in different lost rate condition, then compute PESQ and PESQ-LQO for result file. From following table, we can see that Opus can ’ t congest 20% lost rate as well as SILK. 

Would you please help to check it. 









Regards 

Keith 
_______________________________________________ 
codec mailing list 
[e-mail] 
https://www.ietf.org/mailman/listinfo/codec